actor-agent-frameworks git:20261014.de0823cA

Immutable. This exact content never changes and is served at /api/v1/blob/154964ca442c2d2d.

---
name: actor-agent-frameworks
description: "Implement and manage Actor and Agent Frameworks (e.g. LangGraph). Use when: setup langgraph, implement actor framework, configure agent framework."
license: CC-BY-SA-4.0
metadata:
  source: "agents/skills/ai/actor-agent-frameworks.skill.md"
  tags: "ai"
  core: "false"
  context_cost: "high"
---
# actor-agent-frameworks

## Goal
Implement, configure, and maintain robust Agent and Actor-based frameworks (such as LangGraph, Akka, or custom implementations).

## Steps
1. **Topology Definition**: Define the nodes, edges, and state schema for the agent graph.
2. **Actor Implementation**: Define the explicit behaviors, inbox/message queues, and lifecycle of individual actors.
3. **Integration**: Connect the framework to persistent memory stores (e.g., SQLite, PostgreSQL) for state checkpoints.
4. **Verification**: Simulate actor failures and verify that the graph/supervisor recovers state correctly.

## Security & Guardrails

### 1. Skill Security
- **Sandbox Isolation**: Ensure actors executing untrusted LLM output run in a sandboxed environment without implicit system access.
- **State Poisoning Prevention**: Validate all messages passed between actors against a strict schema (e.g., Pydantic).

### 2. System Integration Security
- **Secure Persistence**: Checkpoints written to the database must not serialize and log plaintext sensitive secrets (API keys).
- **Dead-Letter Handling**: Implement secure dead-letter queues for unhandled messages to prevent memory exhaustion.

### 3. LLM & Agent Guardrails
- **Routing Loop Prevention**: Impose a maximum step/recursion limit on graph traversals to prevent infinite loops.
- **Fail-Safe Constraints**: When an actor fails repeatedly, gracefully escalate to a human operator instead of dropping the context.